Modern high-speed integrated circuit devices, such as synchronous dynamic random access memories (SDRAM), microprocessors, etc., rely upon clock signals to control the flow of commands, data, addresses, etc., into, through, and out of the devices. Additionally, other types of circuit architectures require individual parts to work in unison even though such parts may individually operate at different speeds. As a result, the ability to control the operation of a part through the generation of local clock signals has become increasingly more important. Conventionally, data transfer operations are initiated at the edges of the clock signals (i.e., transitions from high to low or low to high).
In synchronous systems, integrated circuits are synchronized to a common system reference clock. This synchronization often cannot be achieved simply by distributing a single system clock to each of the integrated circuits for the following reason, among others. When an integrated circuit receives a system clock, the circuit often conditions the system clock before the circuit can use the clock. For example, the circuit may buffer the incoming system clock or may convert the incoming system clock from one voltage level to another. This processing introduces its own delay, with the result that the locally processed system clock often will no longer be adequately synchronized with the incoming system clock. The trend toward faster system clock speeds further aggravates this problem since faster clock speeds reduce the amount of delay, or clock skew, which can be tolerated.
To remedy this problem, an additional circuit is conventionally used to synchronize the local clock to the system clock. Two common circuits which are used for this purpose are the phase-locked loop (PLL) and the delay-locked loop (DLL). In the phase-locked loop, a voltage-controlled oscillator produces the local clock. The phases of the local clock and the system clock are compared by a phase-frequency detector, with the resulting error signal used to drive the voltage-controlled oscillator via a loop filter. The feedback via the loop filter phase locks the local clock to the system clock. The delay-locked loop generates a synchronized local clock by delaying the incoming system clock by an integer number of periods. More specifically, the buffers, voltage level converters, etc., of the integrated circuit introduce a certain amount of delay. The delay-locked loop introduces an additional amount of delay such that the resulting local clock is synchronous with the incoming system clock.
In double data rate (DDR) dynamic random access memory (DRAM), wherein operations are initiated on both the rising and the falling edges of the clock signals, it is known to employ a delay lock loop (DDL) to synchronize the output data with the system clock (XCLK) using a phase detector. In an ideal case, the rising edge data is perfectly aligned with the rising edge of the XCLK, the falling edge data is perfectly aligned with the falling edge of the XCLK, and the tAC, or time from when a transition occurs on the XCLK to the time when the data comes through the synchronizing data output (DQ), is within specifications. To approximate an ideal system, a phase detector is conventionally used to lock the rising edge of the DQ signal to the rising edge of the XCLK. In the ideal system, as a result of the rising edge of the DQ signal being phase-locked to the rising edge of the XCLK, the falling edge of the DQ signal changes phase at the same time as the XCLK, or at least within an allowed tolerance (TAC).
A conventional high speed DLL is known to include one negative feedback loop to provide stability and a reliable locking process. Due to the nature of clock synchronization, the accuracy of the DLL over process-voltage-temperature (PVT) differences is strongly dependent on the resemblance between the replica model of the clock path (e.g., clock input buffer, clock mux, clock distribution tree, pre-driver, output driver, etc.) inside of the feedback loop and the actual clock path outside of the DLL.
Unfortunately, however, not all synchronizing circuitry components are “ideal.” Variations in layout, fabrication processes, operating temperatures, and the like, result in non-symmetrical delays among the DLL delay elements. As clock frequency increases, the conventional DLL exhibits an unacceptable tolerance (tAC) variance (i.e., loose distribution) over process-voltage-temperature (PVT) differences. This unacceptable variance over increasing clock frequencies is undesirable for high-speed performance.
It is, therefore, desirable to have synchronizing circuitry including a DLL which compensates for, or at least makes predictable, the variations in delay among the DLL delay elements to enable better matching between the XCLK signal and the DQ signal and thus result in more reliable performance at high speeds.